Research
iph.so explores how neural networks can represent, compress, and predict complex physical and visual processes. Our research focuses on two complementary problems: learning efficient representations of rich data, and learning how those representations evolve over time. From neural video representations to operator learning for fluid transport and scientific simulation, we are interested in the foundations required for AI to model the world rather than merely describe it.
How can neural systems store, reconstruct, and reason about visual information more efficiently than pixels and frames? We study learned representations that enable compact storage, interpolation, and simulation-aware memory.
Nika is our first exploration of this direction.
How can learned models predict the evolution of complex physical systems? We investigate neural operators, fluid transport, convection, and representations that preserve physical structure over long horizons.
Niko applies these ideas to fluid dynamics and scientific machine learning.